home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.databases.informix
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!cis.ohio-state.edu!pacific.mps.ohio-state.edu!linac!att!cbfsb!cbnewsg.cb.att.com!ashaw
- From: ashaw@cbnewsg.cb.att.com (andrew.shaw)
- Subject: Online Performance Degradation
- Message-ID: <1992Sep1.174328.6212@cbfsb.cb.att.com>
- Keywords: select prepare sun
- Sender: news@cbfsb.cb.att.com
- Organization: AT&T
- Date: Tue, 1 Sep 1992 17:43:28 GMT
- Lines: 30
-
- I hope someone can help with a performance mystery with an ESQL-C program.
- We are using Online 4.0 on a Sun 690 w 4 processors, 128M, etc. The
- process starts up and begins doing transactions from a file (adding data,
- with lookups) to an empty database. The performance data looks like this:
-
- Elapsed minutes: 1 2 3 ... 10 ... 35 ... 50
- # of transaction: 220 210 200 150 100 50
-
- If you plot the number of transactions possible against elapsed time you get
- what looks like a log N graph, ie plotted on log paper you get a downward
- sloping straight line. I should point out that since we are doing adds,
- the elapsed time reflects the database getting continuously larger, albeit
- at a slowing rate.
-
- So, the question is: why would performance be falling off in such a drastic
- and sterotyped fashion? The logN nature must surely be a clue, but I can't
- fathom OnLine internals. The performance fall-off slows, but since this
- program is supposed to run briskly over time, I am concerned that the
- performance might be asymptotic to zero transactions per minute.
-
- All of the ESQL-C statements are $prepared the first time they are run,
- and their cursors $declared. There are hardly any joins (if any at all).
- There is a table with a combination (2 ints) index.
-
- Any help would be most appreciated.
-
- Thanks,
- Andrew Shaw
- AT&T Bell Labs
-
-